[llvm] d801b54 - ARM: Fix calling convention for gnu half conversion functions (#147951)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 10 06:47:48 PDT 2025
Author: Matt Arsenault
Date: 2025-07-10T22:47:44+09:00
New Revision: d801b54bcd971f9415da2cce397d12d2129ea1b0
URL: https://github.com/llvm/llvm-project/commit/d801b54bcd971f9415da2cce397d12d2129ea1b0
DIFF: https://github.com/llvm/llvm-project/commit/d801b54bcd971f9415da2cce397d12d2129ea1b0.diff
LOG: ARM: Fix calling convention for gnu half conversion functions (#147951)
I'm surprised at how bad the test coverage is here. There is some
overlap with existing tests, but they aren't comprehensive and do
not cover all the ABIs, or all the different types.
Fixes #147935
Added:
llvm/test/CodeGen/ARM/issue147935-half-convert-libcall-abi.ll
Modified:
llvm/lib/Target/ARM/ARMISelLowering.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 45311e92f9866..cbd43cde78548 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -715,10 +715,14 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM_,
setLibcallImplCallingConv(RTLIB::__truncsfhf2, CallingConv::ARM_AAPCS);
setLibcallImplCallingConv(RTLIB::__truncdfhf2, CallingConv::ARM_AAPCS);
setLibcallImplCallingConv(RTLIB::__extendhfsf2, CallingConv::ARM_AAPCS);
+ setLibcallImplCallingConv(RTLIB::__gnu_h2f_ieee, CallingConv::ARM_AAPCS);
+ setLibcallImplCallingConv(RTLIB::__gnu_f2h_ieee, CallingConv::ARM_AAPCS);
} else {
setLibcallImplCallingConv(RTLIB::__truncsfhf2, CallingConv::ARM_APCS);
setLibcallImplCallingConv(RTLIB::__truncdfhf2, CallingConv::ARM_APCS);
setLibcallImplCallingConv(RTLIB::__extendhfsf2, CallingConv::ARM_APCS);
+ setLibcallImplCallingConv(RTLIB::__gnu_h2f_ieee, CallingConv::ARM_APCS);
+ setLibcallImplCallingConv(RTLIB::__gnu_f2h_ieee, CallingConv::ARM_APCS);
}
}
diff --git a/llvm/test/CodeGen/ARM/issue147935-half-convert-libcall-abi.ll b/llvm/test/CodeGen/ARM/issue147935-half-convert-libcall-abi.ll
new file mode 100644
index 0000000000000..33f3efccd039e
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/issue147935-half-convert-libcall-abi.ll
@@ -0,0 +1,277 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=arm-unknown-linux-gnueabihf -mattr=+vfp2 < %s | FileCheck -check-prefixes=AAPCS-HARD,GNU-AAPCS-HARD %s
+; RUN: llc -mtriple=arm-unknown-linux-gnueabi -mattr=+vfp2 < %s | FileCheck -check-prefixes=AAPCS-SOFT,GNU-AAPCS-SOFT %s
+; RUN: llc -mtriple=arm-unknown-linux-eabihf -mattr=+vfp2 < %s | FileCheck -check-prefixes=AAPCS-HARD,EABI-HARD %s
+; RUN: llc -mtriple=arm-unknown-linux-eabi -mattr=+vfp2 < %s | FileCheck -check-prefixes=AAPCS-SOFT,EABI-SOFT %s
+; RUN: llc -mtriple=armv7-unknown-linux -target-abi=apcs -float-abi=soft < %s | FileCheck -check-prefix=APCS %s
+; RUN: llc -mtriple=armv7-unknown-linux -target-abi=apcs -float-abi=hard < %s | FileCheck -check-prefix=APCS %s
+
+define i1 @test(half %self) #0 {
+; GNU-AAPCS-HARD-LABEL: test:
+; GNU-AAPCS-HARD: @ %bb.0:
+; GNU-AAPCS-HARD-NEXT: .save {r11, lr}
+; GNU-AAPCS-HARD-NEXT: push {r11, lr}
+; GNU-AAPCS-HARD-NEXT: vmov r0, s0
+; GNU-AAPCS-HARD-NEXT: bl __gnu_h2f_ieee
+; GNU-AAPCS-HARD-NEXT: vmov s0, r0
+; GNU-AAPCS-HARD-NEXT: mov r0, #0
+; GNU-AAPCS-HARD-NEXT: vcmp.f32 s0, s0
+; GNU-AAPCS-HARD-NEXT: vmrs APSR_nzcv, fpscr
+; GNU-AAPCS-HARD-NEXT: movvs r0, #1
+; GNU-AAPCS-HARD-NEXT: pop {r11, lr}
+; GNU-AAPCS-HARD-NEXT: mov pc, lr
+;
+; GNU-AAPCS-SOFT-LABEL: test:
+; GNU-AAPCS-SOFT: @ %bb.0:
+; GNU-AAPCS-SOFT-NEXT: .save {r11, lr}
+; GNU-AAPCS-SOFT-NEXT: push {r11, lr}
+; GNU-AAPCS-SOFT-NEXT: bl __gnu_h2f_ieee
+; GNU-AAPCS-SOFT-NEXT: vmov s0, r0
+; GNU-AAPCS-SOFT-NEXT: mov r0, #0
+; GNU-AAPCS-SOFT-NEXT: vcmp.f32 s0, s0
+; GNU-AAPCS-SOFT-NEXT: vmrs APSR_nzcv, fpscr
+; GNU-AAPCS-SOFT-NEXT: movvs r0, #1
+; GNU-AAPCS-SOFT-NEXT: pop {r11, lr}
+; GNU-AAPCS-SOFT-NEXT: mov pc, lr
+;
+; EABI-HARD-LABEL: test:
+; EABI-HARD: @ %bb.0:
+; EABI-HARD-NEXT: .save {r11, lr}
+; EABI-HARD-NEXT: push {r11, lr}
+; EABI-HARD-NEXT: vmov r0, s0
+; EABI-HARD-NEXT: bl __aeabi_h2f
+; EABI-HARD-NEXT: vmov s0, r0
+; EABI-HARD-NEXT: mov r0, #0
+; EABI-HARD-NEXT: vcmp.f32 s0, s0
+; EABI-HARD-NEXT: vmrs APSR_nzcv, fpscr
+; EABI-HARD-NEXT: movvs r0, #1
+; EABI-HARD-NEXT: pop {r11, lr}
+; EABI-HARD-NEXT: mov pc, lr
+;
+; EABI-SOFT-LABEL: test:
+; EABI-SOFT: @ %bb.0:
+; EABI-SOFT-NEXT: .save {r11, lr}
+; EABI-SOFT-NEXT: push {r11, lr}
+; EABI-SOFT-NEXT: bl __aeabi_h2f
+; EABI-SOFT-NEXT: vmov s0, r0
+; EABI-SOFT-NEXT: mov r0, #0
+; EABI-SOFT-NEXT: vcmp.f32 s0, s0
+; EABI-SOFT-NEXT: vmrs APSR_nzcv, fpscr
+; EABI-SOFT-NEXT: movvs r0, #1
+; EABI-SOFT-NEXT: pop {r11, lr}
+; EABI-SOFT-NEXT: mov pc, lr
+;
+; APCS-LABEL: test:
+; APCS: @ %bb.0:
+; APCS-NEXT: push {lr}
+; APCS-NEXT: bl __gnu_h2f_ieee
+; APCS-NEXT: vmov s0, r0
+; APCS-NEXT: mov r0, #0
+; APCS-NEXT: vcmp.f32 s0, s0
+; APCS-NEXT: vmrs APSR_nzcv, fpscr
+; APCS-NEXT: movwvs r0, #1
+; APCS-NEXT: pop {lr}
+; APCS-NEXT: bx lr
+ %_0 = fcmp une half %self, %self
+ ret i1 %_0
+}
+
+define float @f16_to_f32(ptr %p) #0 {
+; GNU-AAPCS-HARD-LABEL: f16_to_f32:
+; GNU-AAPCS-HARD: @ %bb.0:
+; GNU-AAPCS-HARD-NEXT: .save {r11, lr}
+; GNU-AAPCS-HARD-NEXT: push {r11, lr}
+; GNU-AAPCS-HARD-NEXT: ldrh r0, [r0]
+; GNU-AAPCS-HARD-NEXT: bl __gnu_h2f_ieee
+; GNU-AAPCS-HARD-NEXT: vmov s0, r0
+; GNU-AAPCS-HARD-NEXT: pop {r11, lr}
+; GNU-AAPCS-HARD-NEXT: mov pc, lr
+;
+; GNU-AAPCS-SOFT-LABEL: f16_to_f32:
+; GNU-AAPCS-SOFT: @ %bb.0:
+; GNU-AAPCS-SOFT-NEXT: ldrh r0, [r0]
+; GNU-AAPCS-SOFT-NEXT: b __gnu_h2f_ieee
+;
+; EABI-HARD-LABEL: f16_to_f32:
+; EABI-HARD: @ %bb.0:
+; EABI-HARD-NEXT: .save {r11, lr}
+; EABI-HARD-NEXT: push {r11, lr}
+; EABI-HARD-NEXT: ldrh r0, [r0]
+; EABI-HARD-NEXT: bl __aeabi_h2f
+; EABI-HARD-NEXT: vmov s0, r0
+; EABI-HARD-NEXT: pop {r11, lr}
+; EABI-HARD-NEXT: mov pc, lr
+;
+; EABI-SOFT-LABEL: f16_to_f32:
+; EABI-SOFT: @ %bb.0:
+; EABI-SOFT-NEXT: ldrh r0, [r0]
+; EABI-SOFT-NEXT: b __aeabi_h2f
+;
+; APCS-LABEL: f16_to_f32:
+; APCS: @ %bb.0:
+; APCS-NEXT: ldrh r0, [r0]
+; APCS-NEXT: b __gnu_h2f_ieee
+ %load = load half, ptr %p
+ %cvt = fpext half %load to float
+ ret float %cvt
+}
+
+define void @f32_to_f16(ptr %p, float %arg) #0 {
+; GNU-AAPCS-HARD-LABEL: f32_to_f16:
+; GNU-AAPCS-HARD: @ %bb.0:
+; GNU-AAPCS-HARD-NEXT: .save {r4, lr}
+; GNU-AAPCS-HARD-NEXT: push {r4, lr}
+; GNU-AAPCS-HARD-NEXT: mov r4, r0
+; GNU-AAPCS-HARD-NEXT: vmov r0, s0
+; GNU-AAPCS-HARD-NEXT: bl __gnu_f2h_ieee
+; GNU-AAPCS-HARD-NEXT: strh r0, [r4]
+; GNU-AAPCS-HARD-NEXT: pop {r4, lr}
+; GNU-AAPCS-HARD-NEXT: mov pc, lr
+;
+; GNU-AAPCS-SOFT-LABEL: f32_to_f16:
+; GNU-AAPCS-SOFT: @ %bb.0:
+; GNU-AAPCS-SOFT-NEXT: .save {r4, lr}
+; GNU-AAPCS-SOFT-NEXT: push {r4, lr}
+; GNU-AAPCS-SOFT-NEXT: mov r4, r0
+; GNU-AAPCS-SOFT-NEXT: mov r0, r1
+; GNU-AAPCS-SOFT-NEXT: bl __gnu_f2h_ieee
+; GNU-AAPCS-SOFT-NEXT: strh r0, [r4]
+; GNU-AAPCS-SOFT-NEXT: pop {r4, lr}
+; GNU-AAPCS-SOFT-NEXT: mov pc, lr
+;
+; EABI-HARD-LABEL: f32_to_f16:
+; EABI-HARD: @ %bb.0:
+; EABI-HARD-NEXT: .save {r4, lr}
+; EABI-HARD-NEXT: push {r4, lr}
+; EABI-HARD-NEXT: mov r4, r0
+; EABI-HARD-NEXT: vmov r0, s0
+; EABI-HARD-NEXT: bl __aeabi_f2h
+; EABI-HARD-NEXT: strh r0, [r4]
+; EABI-HARD-NEXT: pop {r4, lr}
+; EABI-HARD-NEXT: mov pc, lr
+;
+; EABI-SOFT-LABEL: f32_to_f16:
+; EABI-SOFT: @ %bb.0:
+; EABI-SOFT-NEXT: .save {r4, lr}
+; EABI-SOFT-NEXT: push {r4, lr}
+; EABI-SOFT-NEXT: mov r4, r0
+; EABI-SOFT-NEXT: mov r0, r1
+; EABI-SOFT-NEXT: bl __aeabi_f2h
+; EABI-SOFT-NEXT: strh r0, [r4]
+; EABI-SOFT-NEXT: pop {r4, lr}
+; EABI-SOFT-NEXT: mov pc, lr
+;
+; APCS-LABEL: f32_to_f16:
+; APCS: @ %bb.0:
+; APCS-NEXT: push {r4, lr}
+; APCS-NEXT: mov r4, r0
+; APCS-NEXT: mov r0, r1
+; APCS-NEXT: bl __gnu_f2h_ieee
+; APCS-NEXT: strh r0, [r4]
+; APCS-NEXT: pop {r4, pc}
+ %cvt = fptrunc float %arg to half
+ store half %cvt, ptr %p
+ ret void
+}
+
+define double @f16_to_f64(ptr %p) #0 {
+; GNU-AAPCS-HARD-LABEL: f16_to_f64:
+; GNU-AAPCS-HARD: @ %bb.0:
+; GNU-AAPCS-HARD-NEXT: .save {r11, lr}
+; GNU-AAPCS-HARD-NEXT: push {r11, lr}
+; GNU-AAPCS-HARD-NEXT: ldrh r0, [r0]
+; GNU-AAPCS-HARD-NEXT: bl __gnu_h2f_ieee
+; GNU-AAPCS-HARD-NEXT: vmov s0, r0
+; GNU-AAPCS-HARD-NEXT: vcvt.f64.f32 d0, s0
+; GNU-AAPCS-HARD-NEXT: pop {r11, lr}
+; GNU-AAPCS-HARD-NEXT: mov pc, lr
+;
+; GNU-AAPCS-SOFT-LABEL: f16_to_f64:
+; GNU-AAPCS-SOFT: @ %bb.0:
+; GNU-AAPCS-SOFT-NEXT: .save {r11, lr}
+; GNU-AAPCS-SOFT-NEXT: push {r11, lr}
+; GNU-AAPCS-SOFT-NEXT: ldrh r0, [r0]
+; GNU-AAPCS-SOFT-NEXT: bl __gnu_h2f_ieee
+; GNU-AAPCS-SOFT-NEXT: vmov s0, r0
+; GNU-AAPCS-SOFT-NEXT: vcvt.f64.f32 d0, s0
+; GNU-AAPCS-SOFT-NEXT: vmov r0, r1, d0
+; GNU-AAPCS-SOFT-NEXT: pop {r11, lr}
+; GNU-AAPCS-SOFT-NEXT: mov pc, lr
+;
+; EABI-HARD-LABEL: f16_to_f64:
+; EABI-HARD: @ %bb.0:
+; EABI-HARD-NEXT: .save {r11, lr}
+; EABI-HARD-NEXT: push {r11, lr}
+; EABI-HARD-NEXT: ldrh r0, [r0]
+; EABI-HARD-NEXT: bl __aeabi_h2f
+; EABI-HARD-NEXT: vmov s0, r0
+; EABI-HARD-NEXT: vcvt.f64.f32 d0, s0
+; EABI-HARD-NEXT: pop {r11, lr}
+; EABI-HARD-NEXT: mov pc, lr
+;
+; EABI-SOFT-LABEL: f16_to_f64:
+; EABI-SOFT: @ %bb.0:
+; EABI-SOFT-NEXT: .save {r11, lr}
+; EABI-SOFT-NEXT: push {r11, lr}
+; EABI-SOFT-NEXT: ldrh r0, [r0]
+; EABI-SOFT-NEXT: bl __aeabi_h2f
+; EABI-SOFT-NEXT: vmov s0, r0
+; EABI-SOFT-NEXT: vcvt.f64.f32 d0, s0
+; EABI-SOFT-NEXT: vmov r0, r1, d0
+; EABI-SOFT-NEXT: pop {r11, lr}
+; EABI-SOFT-NEXT: mov pc, lr
+;
+; APCS-LABEL: f16_to_f64:
+; APCS: @ %bb.0:
+; APCS-NEXT: push {lr}
+; APCS-NEXT: ldrh r0, [r0]
+; APCS-NEXT: bl __gnu_h2f_ieee
+; APCS-NEXT: vmov s0, r0
+; APCS-NEXT: vcvt.f64.f32 d16, s0
+; APCS-NEXT: vmov r0, r1, d16
+; APCS-NEXT: pop {lr}
+; APCS-NEXT: bx lr
+ %load = load half, ptr %p
+ %cvt = fpext half %load to double
+ ret double %cvt
+}
+
+define void @f64_to_f16(ptr %p, double %arg) #0 {
+; AAPCS-HARD-LABEL: f64_to_f16:
+; AAPCS-HARD: @ %bb.0:
+; AAPCS-HARD-NEXT: .save {r4, lr}
+; AAPCS-HARD-NEXT: push {r4, lr}
+; AAPCS-HARD-NEXT: mov r4, r0
+; AAPCS-HARD-NEXT: vmov r0, r1, d0
+; AAPCS-HARD-NEXT: bl __aeabi_d2h
+; AAPCS-HARD-NEXT: strh r0, [r4]
+; AAPCS-HARD-NEXT: pop {r4, lr}
+; AAPCS-HARD-NEXT: mov pc, lr
+;
+; AAPCS-SOFT-LABEL: f64_to_f16:
+; AAPCS-SOFT: @ %bb.0:
+; AAPCS-SOFT-NEXT: .save {r4, lr}
+; AAPCS-SOFT-NEXT: push {r4, lr}
+; AAPCS-SOFT-NEXT: mov r4, r0
+; AAPCS-SOFT-NEXT: mov r1, r3
+; AAPCS-SOFT-NEXT: mov r0, r2
+; AAPCS-SOFT-NEXT: bl __aeabi_d2h
+; AAPCS-SOFT-NEXT: strh r0, [r4]
+; AAPCS-SOFT-NEXT: pop {r4, lr}
+; AAPCS-SOFT-NEXT: mov pc, lr
+;
+; APCS-LABEL: f64_to_f16:
+; APCS: @ %bb.0:
+; APCS-NEXT: push {r4, lr}
+; APCS-NEXT: mov r4, r0
+; APCS-NEXT: mov r0, r1
+; APCS-NEXT: mov r1, r2
+; APCS-NEXT: bl __truncdfhf2
+; APCS-NEXT: strh r0, [r4]
+; APCS-NEXT: pop {r4, pc}
+ %cvt = fptrunc double %arg to half
+ store half %cvt, ptr %p
+ ret void
+}
+
+attributes #0 = { nounwind }
More information about the llvm-commits
mailing list